Collision event detection method, apparatus, device, and storage medium

Abstract

The application discloses a vehicle collision event detection method, device and equipment and a storage medium, and relates to the technical field of vehicle collision event detection.The method comprises the following steps: collecting acceleration data and positioning data during vehicle driving; performing noise filtering on the acceleration data to obtain to-be-processed collision data; converting the to-be-processed collision data into frequency domain collision data, and performing initial identification of a collision event in the frequency domain according to the frequency domain collision data; and when a suspected collision event is identified, judging whether a real collision event has occurred according to the to-be-processed collision data and the positioning data.The application judges whether a real collision event has occurred by identifying a collision event according to collected acceleration data and positioning data during vehicle driving, and compared with the prior art method in which a rescue call is used to know that a collision event has occurred, the above-mentioned method can discover a real collision event in time, and rescue efficiency is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of automotive technology, and in particular to a collision event detection method, apparatus, device, and storage medium. Background Technology

[0002] With the rapid development of vehicle networking technology, car manufacturers, 4S stores, and insurance companies have an increasingly strong demand for vehicle collision detection. Usually, insurance companies and 4S stores can only find out that a collision has occurred after receiving a roadside assistance call from a car owner, and then provide services to the car owner. However, such roadside assistance services are often slow.

[0003] First, many new car owners lack experience in handling traffic accidents, are unaware of the procedures, and fail to notify their insurance companies immediately after a collision. Second, in the event of a major accident, or in remote areas where communication is difficult, the car owner may be unable to actively seek help, making it impossible for the service provider to offer professional roadside assistance.

[0004] Therefore, how to promptly detect actual vehicle collisions and effectively improve rescue efficiency has become an urgent problem to be solved.

[0005] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main objective of this invention is to provide a collision event detection method, apparatus, device, and storage medium, aiming to solve the technical problem that the existing technology cannot detect real vehicle collision events in a timely manner, resulting in low rescue efficiency.

[0007] To achieve the above objectives, the present invention provides a collision event detection method, which includes the following steps:

[0008] Collect acceleration and positioning data during vehicle operation;

[0009] The acceleration data is subjected to noise filtering to obtain the collision data to be processed;

[0010] The collision data to be processed is converted into frequency domain collision data, and the initial collision event is identified in the frequency domain based on the frequency domain collision data.

[0011] When a suspected collision event is detected, the system determines whether a real collision event has occurred based on the collision data to be processed and the location data.

[0012] Optionally, the step of filtering noise from the acceleration data to obtain the collision data to be processed includes:

[0013] Filter out data in the acceleration data where the difference between each acceleration is lower than a preset acceleration difference threshold to obtain initial collision data;

[0014] The collision data to be processed is obtained based on the acceleration G value corresponding to the initial collision data.

[0015] Optionally, the step of obtaining the collision data to be processed based on the acceleration G value corresponding to the initial collision data includes:

[0016] The initial collision data is converted into acceleration G-value data;

[0017] Detect the difference in G-values ​​between various acceleration G-value data within a preset time period;

[0018] Delete data in the acceleration G-value data whose G-value difference is lower than a preset G-value difference threshold to obtain collision data to be processed.

[0019] Optionally, the step of converting the collision data to be processed into frequency domain collision data and performing initial collision event identification in the frequency domain based on the frequency domain collision data includes:

[0020] The collision data to be processed is converted into frequency domain collision data, and the natural logarithm value of the frequency band in the frequency domain collision data is obtained in the frequency domain.

[0021] The natural logarithmic value is initially identified for collision events using a preset rectangular sliding window.

[0022] When the natural logarithm values ​​within the window all reach the threshold of the preset rectangular sliding window, a suspected collision event is determined to have occurred.

[0023] Optionally, the step of determining whether a vehicle has actually experienced a collision event based on the collision data to be processed and the positioning data when a suspected collision event is identified includes:

[0024] When a suspected collision event is detected, the collision data to be processed is input into a preset collision event recognition model for recognition.

[0025] If the probability of a collision event is higher than the probability of a non-collision event in the identification results, then the suspected collision event is determined to be a high-probability collision event.

[0026] When the suspected collision event is a high-probability collision event, the non-collision scene information in the suspected collision event is filtered through the positioning data to obtain the target collision result;

[0027] When collision data is detected in the target collision results, it is determined that a real collision has occurred.

[0028] Optionally, before the step of inputting the collision data to be processed into a preset collision event recognition model for recognition when a suspected collision event is identified, the method further includes:

[0029] Obtain the average and standard deviation of acceleration G values ​​from historical collision data during a preset training period;

[0030] Training samples are constructed based on the average value of the acceleration G and the standard deviation of the acceleration G;

[0031] The weights of the training samples are adjusted based on a preset loss function to obtain the target training samples;

[0032] The initial linear model is iteratively trained using the target training samples to obtain a preset collision event recognition model.

[0033] Optionally, after determining whether a real collision event has occurred based on the collision data to be processed and the positioning data when a suspected collision event is identified, the method further includes:

[0034] When a vehicle is determined to have experienced a real collision, the actual collision time is obtained, and the actual acceleration data and actual positioning data before and after the actual collision time are recorded.

[0035] A collision alarm is triggered based on the actual acceleration data and the actual positioning data.

[0036] The collision alarm information is sent to the backend server so that the backend server can perform accident handling operations.

[0037] Furthermore, to achieve the above objectives, the present invention also proposes a collision event detection device, the device comprising:

[0038] The data acquisition module is used to collect acceleration and positioning data during vehicle operation.

[0039] A noise filtering module is used to filter noise from the acceleration data to obtain collision data to be processed.

[0040] The collision recognition module is used to convert the collision data to be processed into frequency domain collision data, and to perform initial collision event recognition in the frequency domain based on the frequency domain collision data.

[0041] The collision determination module is used to determine whether a vehicle has actually experienced a collision event when a suspected collision event is detected, based on the collision data to be processed and the positioning data.

[0042] Furthermore, to achieve the above objectives, the present invention also proposes a collision event detection device, the device comprising: a memory, a processor, and a collision event detection program stored in the memory and executable on the processor, the collision event detection program being configured to implement the steps of the collision event detection method as described above.

[0043] In addition, to achieve the above objectives, the present invention also proposes a storage medium storing a collision event detection program, which, when executed by a processor, implements the steps of the collision event detection method as described above.

[0044] This invention collects acceleration and positioning data during vehicle movement, filters noise from the acceleration data to obtain collision data to be processed, converts this collision data into frequency domain collision data, and performs initial collision event identification based on the frequency domain collision data. Finally, when a suspected collision event is identified, it determines whether a real collision event has occurred based on the collision data to be processed and the positioning data. Because this invention automatically identifies collision events by collecting acceleration and positioning data during vehicle movement to determine whether a real collision event has occurred, compared to existing methods that only know a collision has occurred after receiving a roadside assistance call from the owner and then provide service, this invention can promptly detect real collision events, effectively improving rescue efficiency. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the structure of a collision event detection device in the hardware operating environment involved in the embodiments of the present invention;

[0046] Figure 2 This is a flowchart illustrating the first embodiment of the collision event detection method of the present invention;

[0047] Figure 3 This is a graph showing the relationship between acceleration values ​​and time along the x, y, and z axes in the first embodiment of the present invention.

[0048] Figure 4 This is a flowchart illustrating the second embodiment of the collision event detection method of the present invention;

[0049] Figure 5 This is a graph showing the relationship between acceleration G and time in the second embodiment of the present invention;

[0050] Figure 6 This is an image showing the relationship between collision data to be processed and time in the second embodiment of the present invention;

[0051] Figure 7This is a graph showing the relationship between frequency domain collision data and frequency in the second embodiment of the present invention;

[0052] Figure 8 This is an image of the process of recognizing natural logarithmic values ​​using a preset rectangular sliding window in the second embodiment of the present invention;

[0053] Figure 9 This is a flowchart illustrating the third embodiment of the collision event detection method of the present invention;

[0054] Figure 10 This is a structural block diagram of the first embodiment of the collision event detection device of the present invention.

[0055] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0056] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0057] Reference Figure 1 , Figure 1 This is a schematic diagram of the collision event detection device structure in the hardware operating environment involved in the embodiments of the present invention.

[0058] like Figure 1 As shown, the collision event detection device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0059] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the collision event detection device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0060] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a collision event detection program.

[0061] exist Figure 1 In the collision event detection device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the collision event detection device of the present invention can be set in the collision event detection device, and the collision event detection device calls the collision event detection program stored in the memory 1005 through the processor 1001 and executes the collision event detection method provided in the embodiment of the present invention.

[0062] This invention provides a collision event detection method, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the collision event detection method of the present invention.

[0063] In this embodiment, the collision event detection method includes the following steps:

[0064] Step S10: Collect acceleration and positioning data during vehicle operation.

[0065] It should be noted that the executing entity of the method in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a mobile phone, tablet computer, or personal computer, or other electronic devices capable of performing the same or similar functions. The following description uses the aforementioned collision event detection device (hereinafter referred to as the detection device) to illustrate this embodiment and the subsequent embodiments.

[0066] Understandably, when a vehicle is involved in a collision, its acceleration data during its journey will change significantly. Therefore, acceleration data can be used to determine whether a real collision has occurred.

[0067] It should be noted that the detection equipment can collect acceleration and positioning data during vehicle operation through acceleration and positioning sensors on the vehicle-mounted equipment.

[0068] It should be noted that acceleration data can be collected by an acceleration sensor in the vehicle, such as a common three-axis G-sensor sensor, which can record acceleration along three different axes (x-axis, y-axis, and z-axis) during vehicle movement.

[0069] Understandably, positioning data can be collected by positioning sensors in vehicle-mounted equipment, including GNSS positioning sensors. GNSS is a global navigation satellite system, which functions similarly to GPS, providing versatile, all-weather, continuous, and real-time navigation, positioning, and timing capabilities. It can provide users with precise three-dimensional coordinates, velocity, and time. Furthermore, the GNSS positioning sensor can simultaneously receive GPS / BeiDou dual-mode transmissions to obtain more accurate positioning data.

[0070] In practice, the detection device can adjust the reporting frequency of the acceleration sensor (e.g., 100Hz) to record the acceleration values ​​on the x, y, and z axes every minute. The recorded acceleration values ​​are used as acceleration data, and the positioning data during vehicle movement can be obtained through the positioning sensor.

[0071] For ease of understanding, please refer to Figure 3 This explanation does not limit the scope of this solution. Figure 3 This is a graph showing the relationship between acceleration values ​​on the x, y, and z axes and time in the first embodiment of the present invention. The graph records the relationship between the components of acceleration data on the x, y, and z axes and time within one minute; the graphs for other time periods are similar. Due to the Earth's gravity, under normal installation conditions, the device's acceleration and gravitational acceleration magnitude are such that a gravitational acceleration of 1g can be sensed in the z-axis direction, the x-axis data is essentially close to 0, and the y-axis has a small component. If the device is moved or tilted, the vertical gravitational acceleration will be decomposed into the x, y, and z axes.

[0072] Step S20: Filter the acceleration data for noise to obtain collision data to be processed.

[0073] It should be noted that the raw data collected from the accelerometer is very messy and has a lot of noise. Judging the actual collision based solely on the data collected from the accelerometer is not very accurate. Therefore, it is necessary to filter the acceleration data to improve the accuracy of judging the actual collision.

[0074] Understandably, noise can be messy and useless interference data in acceleration data, which can affect the accuracy of actual collision judgment, and therefore needs to be filtered out.

[0075] It should be noted that the collision data to be processed can be acceleration data after filtering out noise.

[0076] In practice, the detection device can further reduce noise by deleting points where the difference between acceleration data is below a preset difference threshold, thereby filtering out data with too small a difference between acceleration data.

[0077] It should be understood that the preset difference threshold can be determined by the difference between actual accelerations during past vehicle driving.

[0078] Step S30: Convert the collision data to be processed into frequency domain collision data, and perform initial collision event identification in the frequency domain based on the frequency domain collision data.

[0079] It should be noted that the collision data to be processed is acceleration data at various times, which cannot determine the frequency distribution of collision events. Therefore, it is necessary to convert the above collision data to be processed into data in the frequency domain, i.e., frequency domain collision data, in order to identify collision events.

[0080] In practical implementation, the detection device can perform initial collision event identification on the frequency domain collision data through pattern recognition in the frequency domain. When the value of the frequency domain collision data within a certain bandwidth is greater than the preset identification threshold, it is determined that the vehicle has experienced a suspected collision event; otherwise, the vehicle is in a normal state.

[0081] It should be understood that pattern recognition can be the classification of samples based on their features using computer calculations, and the preset recognition threshold can be the benchmark value used to classify samples in pattern recognition.

[0082] Step S40: When a suspected collision event is identified, determine whether a real collision event has occurred based on the collision data to be processed and the positioning data.

[0083] In practical implementation, when the detection device identifies a suspected collision event, it can filter the data in the collision data to be processed that corresponds to non-collision events to increase the probability that the suspected collision event is a real collision event. Then, it can fuse the positioning data to filter out the data in the collision data to be processed that corresponds to non-collision scenarios, thereby obtaining highly accurate target collision data. By detecting the target collision data, the vehicle algorithm can determine whether a real collision event has occurred.

[0084] It should be understood that the target collision data obtained after the above processing is data corresponding to high-accuracy collision events. In other words, if the target collision data contains data, it can be determined that a real collision event has occurred; otherwise, it can be determined that a real collision event has not occurred.

[0085] This embodiment collects acceleration and positioning data during vehicle movement, filters the acceleration data for noise to obtain collision data to be processed, converts this collision data into frequency domain collision data, and performs initial collision event identification based on the frequency domain collision data. Finally, when a suspected collision event is identified, it determines whether a real collision event has occurred based on the collision data to be processed and the positioning data. Because this embodiment automatically identifies collision events using the collected acceleration and positioning data during vehicle movement to determine whether a real collision event has occurred, compared to existing technologies that only know a collision has occurred after receiving a roadside assistance call from the owner and then provide service, this embodiment can promptly detect real collision events, effectively improving rescue efficiency.

[0086] Furthermore, in order to provide timely rescue services to vehicle owners involved in actual collisions, this embodiment further includes the following step after step S40:

[0087] Step S50: When it is determined that a real collision event has occurred, the actual collision time is obtained, and the actual acceleration data and actual positioning data before and after the actual collision time are recorded.

[0088] It should be noted that the actual collision time can be the duration of the actual collision, which can be recorded by the positioning sensor.

[0089] In practice, when a vehicle is determined to have experienced a real collision, the detection equipment can collect actual acceleration data and actual positioning data before and after the actual collision time using acceleration and positioning sensors.

[0090] Step S60: Trigger a collision alarm based on the actual acceleration data and the actual positioning data.

[0091] It should be noted that collision warning information can be used to alert the vehicle to a real collision.

[0092] In a practical implementation, the detection device can convert the actual acceleration data and the actual positioning data into information for subsequent transmission to the backend server.

[0093] Step S70: Send the collision alarm information to the backend server so that the backend server can perform accident handling operations.

[0094] It should be noted that the backend server can be a server used for vehicle collision insurance information. This backend server can be an operable frontend or a backend that connects to other roadside assistance service providers. If it is a backend, it can be a frontend.

[0095] In practice, the detection equipment can transmit the aforementioned collision alarm information to the backend server via a wireless communication module. If the backend server is an operable frontend, the collision alarm information is saved to the database. The frontend page used by customer service personnel will then notify them that a real collision event has occurred. Customer service personnel can retrieve the corresponding vehicle's contact information from the system to determine how to contact the vehicle owner promptly and provide guidance on handling the collision or roadside assistance services.

[0096] Furthermore, if the backend server is connected to the backend of other rescue service providers, the collision alarm information will be sent to the corresponding rescue service provider.

[0097] It should be understood that the aforementioned wireless communication module may be a module installed inside the detection equipment for communicating with the outside, such as WIFI.

[0098] This embodiment triggers a collision alarm based on acceleration and positioning data before and after the collision when a real collision event is detected. The collision alarm is then sent to a backend server, which performs accident handling operations. This allows for timely rescue services to vehicle owners involved in real collision events, effectively improving their driving experience.

[0099] refer to Figure 4 , Figure 4 This is a flowchart illustrating the second embodiment of the collision event detection method of the present invention.

[0100] Based on the first embodiment described above, in this embodiment, step S20 includes:

[0101] Step S201: Filter the data in the acceleration data where the difference between each acceleration is lower than a preset acceleration difference threshold to obtain initial collision data.

[0102] It should be noted that directly filtering the acceleration data for noise results in poor filtering effectiveness. Therefore, in order to improve the filtering effect and obtain more accurate data, a second embodiment of the present invention is proposed.

[0103] Understandably, the preset acceleration difference threshold can be a value used to filter acceleration data, which can be determined by the acceleration function relationship or by the actual acceleration of passing vehicles during their travel.

[0104] It should be noted that the initial collision data can be acceleration data filtered after passing through a preset acceleration difference threshold.

[0105] In the specific implementation, the detection device detects the absolute value of the difference between the acceleration data at each point on the x, y, and z axes. If the absolute value is greater than the preset acceleration difference threshold (e.g., 0.3g), it is recorded as a collision point. Then, the x, y, and z data within each preset recording time period (e.g., 30s) before and after the collision point are obtained as the initial collision data.

[0106] It should be understood that the above-mentioned preset recording period can be adjusted according to actual needs, and this embodiment does not limit it in this way.

[0107] Step S202: Obtain the collision data to be processed based on the acceleration G value corresponding to the initial collision data.

[0108] It should be noted that the acceleration G value is the value obtained by dividing the acceleration value within a certain acceleration range by the gravitational acceleration G. For example, if a certain model of car has an acceleration time of 6.2s in a 0-100km / h acceleration performance test, then its average acceleration is 27.78 / 6.2 = 4.48m / s². Therefore, the acceleration G value is 4.48 / 9.8 + 0.46, which is 0.46g.

[0109] Understandably, the collision data to be processed can be G-value data after further filtering of the above acceleration G-values.

[0110] For ease of understanding, please refer to Figure 5 This explanation does not limit the scope of this solution. Figure 5 This is a graph showing the relationship between acceleration G and time in the second embodiment of the present invention. In the graph, the acceleration values ​​of x, y, and z in the collision data to be processed are combined into the relationship between G and time.

[0111] Furthermore, in order to more accurately improve the filtration effect, in this embodiment, step S202 includes:

[0112] Step S2021: Convert the initial collision data into acceleration G value data.

[0113] In practical implementation, the detection equipment can be... The x, y, and z acceleration values ​​in the collision data to be processed are combined into a G value. However, the |G| obtained in the above way is a scalar and cannot reflect the angle change of the synthesized G value. Further, the magnitude of the vector product of the zero-point G value and the current G value can be used as the actual G value, which can simultaneously measure the magnitude and direction change of G.

[0114] It should be understood that the aforementioned zero-point G value can be determined from the first few data points in the collision data to be processed. If the G values ​​corresponding to the first few data points are basically consistent, then these data points can be considered as a stationary zero point, and the vector G of this point can be taken as the zero-point vector G. The magnitude of this zero-point vector G is the aforementioned zero-point G value.

[0115] Step S2022: Detect the G-value difference between the acceleration G-value data within the preset time period.

[0116] It should be noted that the preset time period can be the time period selected based on the detection. This embodiment takes one minute as an example, but it does not limit this solution.

[0117] In practice, the detection device can detect the difference in G-values ​​between various acceleration G-value data within one minute to filter out points with small differences.

[0118] Step S2023: Delete data in the acceleration G-value data whose G-value difference is lower than a preset G-value difference threshold to obtain collision data to be processed.

[0119] It should be noted that the preset G-value difference threshold can be a value used to filter acceleration G-value data. It can be determined by the acceleration G-value function relationship or by the actual acceleration G-value data of past vehicles.

[0120] In practice, the detection device can filter out acceleration G-value data whose G-value difference is lower than a preset G-value difference threshold, and use the remaining acceleration G-value data as collision data to be processed.

[0121] It should be noted that this embodiment improves the noise filtering effect by performing preliminary filtering on the acceleration data and then filtering the acceleration G value corresponding to the initial collision data obtained by filtering again, thus obtaining collision data to be processed with higher accuracy.

[0122] Furthermore, to improve the accuracy of initial collision event identification, in this embodiment, step S30 includes:

[0123] Step S301: Convert the collision data to be processed into frequency domain collision data, and obtain the natural logarithm value of the frequency band in the frequency domain collision data.

[0124] For ease of understanding, please refer to Figure 6 This explanation does not limit the scope of this solution. Figure 6 The image shows the relationship between the collision data to be processed and time in the second embodiment of the present invention. The image records the relationship between the acceleration G value in the collision data to be processed and time. It can be seen that the collision data to be processed is the acceleration data at each moment. It is impossible to determine the frequency distribution of the collision event. Therefore, it is necessary to convert the value for identification in the frequency domain.

[0125] It should be noted that the collision data to be processed can be converted from the time domain to the frequency domain using FFT (Fast Fourier Transform) to obtain frequency domain collision data.

[0126] For ease of understanding, please refer to Figure 7 This explanation does not limit the scope of this solution. Figure 7 This is a graph showing the relationship between frequency domain collision data and frequency in the second embodiment of the present invention. The graph records the relationship between the numerical value of the frequency domain collision data and the frequency domain.

[0127] In practice, to reduce the absolute value of the frequency domain collision data and facilitate subsequent identification, the natural logarithmic value of each frequency band value of the frequency domain collision data can be taken.

[0128] For ease of understanding, please refer to Figure 8 This explanation does not limit the scope of this solution. Figure 8 The image shows the process of recognizing natural logarithmic values ​​using a preset rectangular sliding window in the second embodiment of the present invention. In the image, the rectangular part is ignored, and the remaining curved part is the relationship between the natural logarithmic value and the frequency of each frequency band of the frequency domain collision data.

[0129] Step S302: Initially identify collision events for the natural logarithmic value using a preset rectangular sliding window.

[0130] It should be noted that the preset rectangular sliding window can be a rectangular template used to classify the features of the aforementioned natural logarithmic values, such as... Figure 8 As shown in the figure, the rectangular part is the preset rectangular sliding window. In this embodiment, the preset rectangular sliding window with two sets of thresholds (high and low) is used for illustration. However, in the actual recognition process, one set of thresholds or multiple sets of thresholds can also be used. This embodiment does not limit this.

[0131] Step S303: When the natural logarithm values ​​within the window all reach the threshold of the preset rectangular sliding window, it is determined that a suspected collision event has occurred.

[0132] It should be noted that, as Figure 8 As shown, the threshold of the preset rectangular sliding window is the value corresponding to the width of the rectangle. In the figure, the left-hand rectangular portion has a higher threshold, indicating high sensitivity (8Hz bandwidth, dB>2.8), while the right-hand rectangular portion has low sensitivity (10Hz bandwidth, dB>3.0). The bandwidth and dB values ​​can be changed according to actual needs; this embodiment does not impose any restrictions on this.

[0133] In a specific implementation, if the natural logarithmic values ​​within the window all reach the threshold of the preset rectangular sliding window, it can be determined that a suspected collision event has occurred; otherwise, it can be determined that no collision event has occurred.

[0134] It should be understood that this embodiment uses a preset rectangular sliding window to identify collision events in the frequency domain of the collision data to be processed, which effectively improves the accuracy of collision event identification.

[0135] refer to Figure 9 , Figure 9 This is a flowchart illustrating the third embodiment of the collision event detection method of the present invention.

[0136] Based on the above embodiments, in this embodiment, step S40 includes:

[0137] Step S401: When a suspected collision event is identified, the collision data to be processed is input into a preset collision event recognition model for recognition.

[0138] It should be noted that determining whether a sample is a real collision is a binary classification problem encountered in AI. Using an AI intelligent model to solve the classification can make the probability of a suspected collision event being a real collision event more accurate, thereby improving the accuracy of the determination of real collision events.

[0139] Understandably, the preset collision event recognition model can be an AI model trained based on previous collision data, used to identify the probability of real collision events in the collision data to be processed.

[0140] In the specific implementation, when a suspected collision time is identified, the frequency domain collision data is converted from the frequency domain to the time domain to obtain the collision data to be processed. The collision data to be processed is divided into time periods, such as every two seconds as a time period. The average G value of each time period is recorded as an item, such as g0, g1, g2...g_last. The standard deviation of the G value of each time period is recorded as an item, such as std0, std1, std2...std_last. At the same time, the average G value and standard deviation of the G value of the samples in each stage are recorded to a CSV file. The samples are input into the preset collision event recognition model through the CSV file. The preset collision event recognition model loads the sample data from the CSV file for recognition.

[0141] Step S402: If the probability of a collision event in the identification result is higher than the probability of a non-collision event, then the suspected collision event is determined to be a high-probability collision event.

[0142] In the specific implementation, the preset collision event recognition model outputs only two types of results: the probability of a collision event and the probability of a non-collision event. If the probability of a collision event is higher than the probability of a non-collision event in the recognition result, the suspected collision event is determined to be a high-probability collision event; otherwise, the suspected collision event is determined to be a low-probability collision event.

[0143] Step S403: When the suspected collision event is a high-probability collision event, filter the non-collision scene information in the suspected collision event through the positioning data to obtain the target collision result.

[0144] It should be noted that the target collision result can be the collision data in the collision data to be processed after filtering out non-collision scene information data.

[0145] It should be noted that non-collision scene information can be actual scene information that is misjudged as a collision event. The recognition results output by the above-mentioned preset collision event recognition model will be relatively poor in the above-mentioned non-collision scene. Therefore, it is necessary to filter the above-mentioned non-collision scene information through positioning data.

[0146] Understandably, when a vehicle is stationary, a collision is almost impossible. The most common scenarios are washing the car or hitting the windshield, which correspond to the non-collision scenarios mentioned above.

[0147] It should be understood that if a vehicle is traveling at a constant speed and a collision occurs at high speed, the driver will inevitably brake or the vehicle will deviate from its original direction of travel. Therefore, in cases where a collision has been triggered but the vehicle continues to travel at high speed, the vast majority of the time it is due to severe bumps on a poor road surface, the recording device being dropped, or the owner intentionally hitting the recording device. These situations also correspond to the scenarios mentioned above for non-collision scenarios.

[0148] In a specific implementation, the detection device filters the non-collision scene information in the suspected collision event by fusing positioning data to obtain a more accurate target collision result. In other words, by fusing positioning data to filter out non-collision scene data in the collision data to be processed, the remaining collision data is used as the target collision result.

[0149] Step S404: When collision data is detected in the target collision results, it is determined that a real collision has occurred.

[0150] In the specific implementation, if the non-collision scene data in the collision data to be processed is filtered out by the positioning data, then there is no collision data in the target collision result. That is to say, the original data in the collision data to be processed are all non-collision scene data, and the vehicle has not actually collided. Otherwise, it means that there is collision data in the target collision result, and the collision data is not non-collision data, and the vehicle has actually collided.

[0151] This embodiment uses a preset collision event recognition model to further identify vehicle collision data that is determined to be a suspected collision event, and then filters the recognition results with positioning data to obtain a highly accurate target collision result. Based on the target collision result, it is determined whether a real collision exists, which effectively improves the accuracy of determining real collision events.

[0152] Furthermore, to improve the accuracy of the preset collision event recognition model, in this embodiment, before step S401, the following steps are also included:

[0153] Step S4011: Obtain the average value and standard deviation of acceleration G value of historical collision data under the preset training period.

[0154] It should be noted that historical collision data can be any collision data to be processed obtained in the past using the methods described above.

[0155] Understandably, the preset training period is used to determine the average value and standard deviation of the acceleration G value, and can be adjusted according to actual needs.

[0156] In practice, the detection device can construct training samples by using the average value and standard deviation of acceleration G values ​​from historical collision data during a preset training period.

[0157] Step S4012: Construct training samples based on the average value of the acceleration G and the standard deviation of the acceleration G.

[0158] In practice, the detection device divides the average acceleration G-value and acceleration G-value of a portion of the historical collision data into a training set and the average acceleration G-value and acceleration G-value of another portion of the data into a test set. For example, the average acceleration G-value and acceleration G-value of 4,000 collision data from the second half of 2021 (6 months) are used as the training set, and the average acceleration G-value and acceleration G-value of 1,000 collision data from January to February 2022 are used as the test set to construct training samples for training.

[0159] Step S4013: Adjust the weights of the training samples based on a preset loss function to obtain the target training samples.

[0160] It should be noted that the preset loss function can be a function used to balance the proportion of positive and negative samples in the training samples.

[0161] Understandably, since the historical collision data is the aforementioned collision data to be processed, and the actual collisions obtained through the above method are relatively few, accounting for only about 1%, directly training from the above training samples will lead to the problem of program sample imbalance. Therefore, it is necessary to adjust the training samples through a preset loss function.

[0162] It should be noted that commonly used functions in binary cross-entropy training, such as binary_crossentropy, calculate too many negative sample results, masking the positive sample results and making the positive sample results insignificant in training. Therefore, it is necessary to adjust the sample weights using the aforementioned preset loss function, such as focal.

[0163] If the focal function described above is used as the loss function, then the focal function can be:

[0164]

[0165] make

[0166] In practice, since the number of positive samples is much lower than that of negative samples, and positive samples are difficult to classify, it is necessary to take an alpha value that is as large as possible. When the alpha value is relatively large, the calculation result of whether the sample is identified as a positive sample will have a higher weight in the loss function, which can bring better accuracy to the identification result.

[0167] Understandably, imbalanced samples can also be addressed through scaling strategies, i.e., adjusting the classification threshold of a preset rectangular sliding window to move the threshold towards a smaller value, thereby more closely approximating the proportion of the true samples.

[0168] Step S4014: Iteratively train the initial linear model using the target training samples to obtain a preset collision event recognition model.

[0169] It should be noted that the initial linear model can be a model with a learning algorithm, such as Sequential Mode 1, which uses a fully connected intermediate layer and a fully connected output layer with a sigmoid activation function.

[0170] This embodiment adjusts the weights of training samples by using a preset loss function to balance the proportion of positive and negative samples. Then, iterative training of the initial linear model is performed using the adjusted target training samples to obtain a preset event recognition model, which effectively improves the accuracy of the preset collision event recognition model.

[0171] Furthermore, embodiments of the present invention also propose a storage medium storing a collision event detection program, wherein the collision event detection program, when executed by a processor, implements the steps of the collision event detection method described above.

[0172] Reference Figure 5 , Figure 10 This is a structural block diagram of the first embodiment of the collision event detection device of the present invention.

[0173] like Figure 10 As shown, the collision event detection device proposed in this embodiment of the invention includes:

[0174] The data acquisition module 501 is used to collect acceleration data and positioning data during vehicle operation;

[0175] Noise filtering module 502 is used to filter noise from the acceleration data to obtain collision data to be processed;

[0176] The collision recognition module 503 is used to convert the collision data to be processed into frequency domain collision data, and to perform initial collision event recognition in the frequency domain based on the frequency domain collision data.

[0177] The collision determination module 504 is used to determine whether a real collision event has occurred by the vehicle based on the collision data to be processed and the positioning data when a suspected collision event is identified.

[0178] This embodiment collects acceleration and positioning data during vehicle movement, filters the acceleration data for noise to obtain collision data to be processed, converts this collision data into frequency domain collision data, and performs initial collision event identification based on the frequency domain collision data. Finally, when a suspected collision event is identified, it determines whether a real collision event has occurred based on the collision data to be processed and the positioning data. Because this embodiment automatically identifies collision events using the collected acceleration and positioning data during vehicle movement to determine whether a real collision event has occurred, compared to existing technologies that only know a collision has occurred after receiving a roadside assistance call from the owner and then provide service, this embodiment can promptly detect real collision events, effectively improving rescue efficiency.

[0179] Other embodiments or specific implementations of the collision event detection device of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0180] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0181] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0182] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0183] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for detecting vehicle collision events, characterized in that, The vehicle collision event detection method includes the following steps: Collect acceleration and positioning data during vehicle operation; The acceleration data is subjected to noise filtering to obtain the collision data to be processed; The collision data to be processed is converted into frequency domain collision data, and the initial collision event is identified in the frequency domain based on the frequency domain collision data. When a suspected collision event is detected, it is determined whether a real collision event has occurred based on the collision data to be processed and the positioning data. The step of converting the collision data to be processed into frequency domain collision data and performing initial collision event identification in the frequency domain based on the frequency domain collision data includes: The collision data to be processed is transformed from the time domain to the frequency domain by using a fast Fourier transform to obtain frequency domain collision data; Obtain the natural logarithm value of the frequency band in the frequency domain collision data; The natural logarithmic value is initially identified for collision events using a preset rectangular sliding window. When the natural logarithm values ​​within the window all reach the threshold of the preset rectangular sliding window, a suspected collision event is determined to have occurred.

2. The collision event detection method as described in claim 1, characterized in that, The step of filtering noise from the acceleration data to obtain the collision data to be processed includes: Filter out data in the acceleration data where the difference between each acceleration is lower than a preset acceleration difference threshold to obtain initial collision data; The collision data to be processed is obtained based on the acceleration G value corresponding to the initial collision data.

3. The collision event detection method as described in claim 2, characterized in that, The step of obtaining the collision data to be processed based on the acceleration G value corresponding to the initial collision data includes: The initial collision data is converted into acceleration G-value data; Detect the difference in G-values ​​between various acceleration G-value data within a preset time period; Delete data in the acceleration G-value data whose G-value difference is lower than a preset G-value difference threshold to obtain collision data to be processed.

4. The collision event detection method as described in claim 1, characterized in that, The step of determining whether a vehicle has actually collided based on the collision data to be processed and the positioning data when a suspected collision event is identified includes: When a suspected collision event is detected, the collision data to be processed is input into a preset collision event recognition model for recognition. If the probability of a collision event is higher than the probability of a non-collision event in the identification results, then the suspected collision event is determined to be a high-probability collision event. When the suspected collision event is a high-probability collision event, the non-collision scene information in the suspected collision event is filtered through the positioning data to obtain the target collision result; When collision data is detected in the target collision results, it is determined that a real collision has occurred.

5. The collision event detection method as described in claim 4, characterized in that, Before the step of inputting the collision data to be processed into a preset collision event recognition model for recognition when a suspected collision event is identified, the method further includes: Obtain the average and standard deviation of acceleration G values ​​from historical collision data during a preset training period; Training samples are constructed based on the average value of the acceleration G and the standard deviation of the acceleration G. The weights of the training samples are adjusted based on a preset loss function to obtain the target training samples; The initial linear model is iteratively trained using the target training samples to obtain a preset collision event recognition model.

6. The collision event detection method according to any one of claims 1 to 5, characterized in that, After the step of determining whether a vehicle has actually experienced a collision based on the collision data to be processed and the positioning data when a suspected collision event is identified, the method further includes: When a vehicle is determined to have experienced a real collision, the actual collision time is obtained, and the actual acceleration data and actual positioning data before and after the actual collision time are recorded. A collision alarm is triggered based on the actual acceleration data and the actual positioning data. The collision alarm information is sent to the backend server so that the backend server can perform accident handling operations.

7. A collision event detection device, characterized in that, The device includes: The data acquisition module is used to collect acceleration and positioning data during vehicle operation. A noise filtering module is used to filter noise from the acceleration data to obtain collision data to be processed. The collision recognition module is used to convert the collision data to be processed into frequency domain collision data, and to perform initial collision event recognition in the frequency domain based on the frequency domain collision data. The collision determination module is used to determine whether a real collision event has occurred by the vehicle based on the collision data to be processed and the positioning data when a suspected collision event is identified. The collision recognition module is also used for: The collision data to be processed is transformed from the time domain to the frequency domain by using a fast Fourier transform to obtain frequency domain collision data; Obtain the natural logarithm value of the frequency band in the frequency domain collision data; The natural logarithmic value is initially identified for collision events using a preset rectangular sliding window. When the natural logarithm values ​​within the window all reach the threshold of the preset rectangular sliding window, a suspected collision event is determined to have occurred.

8. A collision event detection device, characterized in that, The device includes: a memory, a processor, and a collision event detection program stored in the memory and executable on the processor, the collision event detection program being configured to implement the steps of the collision event detection method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a collision event detection program, which, when executed by a processor, implements the steps of the collision event detection method as described in any one of claims 1 to 6.

Images